|Número de publicación||US7589648 B1|
|Tipo de publicación||Concesión|
|Número de solicitud||US 11/054,855|
|Fecha de publicación||15 Sep 2009|
|Fecha de presentación||10 Feb 2005|
|Fecha de prioridad||10 Feb 2005|
|Número de publicación||054855, 11054855, US 7589648 B1, US 7589648B1, US-B1-7589648, US7589648 B1, US7589648B1|
|Inventores||Benny Ma, San-Ta Kow, Ann Wu, Thomas Tsui|
|Cesionario original||Lattice Semiconductor Corporation|
|Exportar cita||BiBTeX, EndNote, RefMan|
|Citas de patentes (38), Otras citas (8), Citada por (8), Clasificaciones (11), Eventos legales (4)|
|Enlaces externos: USPTO, Cesión de USPTO, Espacenet|
The present invention relates generally to data decompression, and more particularly to the dynamic decompression of data.
Many types of electronic devices require that information be loaded from a source and stored in a memory. For example, programmable logic devices are configured to implement a desired logical function based upon configuration data provided by a programming tool. The configuration data may be stored internally for a non-volatile device or externally for a volatile device. Regardless of whether the configuration data is stored internally or externally, the increasing complexity of programmable logic devices requires larger and larger amounts of configuration data. This increased configuration data size produces delays in the configuration process and increases the costs.
During the configuration process, the configuration data is typically loaded from a source such as an EEPROM into a programmable logic device responsive to cycles of a clock signal. To reduce the storage requirement for the EEPROM, the configuration data may be compressed. In addition, the compression of the configuration data decreases the amount of time needed to configure the corresponding programmable logic device. Because a programmable logic device is very sensitive to errors in the configuration data stream, any compression of the data must be lossless or perfect such that the decompressed configuration data is exactly the same as the configuration data before compression. In addition, only a portion of the configuration data can generally be compressed in light of the requirement for perfect compression. Thus, the configuration data being shifted into the device will comprise both uncompressed and compressed data. This mixed nature of the configuration data complicates the configuration data flow. Any compressed configuration data must be decompressed before configuration may be completed. However, the configuration data is being shifted at a constant rate responsive to cycles of the clock signal. When data is decompressed, this constant rate must necessarily change. This adds complexity to the configuration process because it must accommodate the rate changes depending upon whether configuration data being shifted in was compressed or not. This accommodation may take place in the programming tool providing the bit stream or may take place within the device. Regardless of where the accommodation takes place, it adds considerable complexity to the configuration circuitry and the software controlling the configuration process. Moreover, this complexity inherently limits the bit rate of the configuration data stream. As a result of this complexity, comparatively few programmable devices utilize configuration data compression, particularly in low-cost devices.
Accordingly, there is a need in the art for improved decompression techniques for data streams.
One aspect of the invention relates to a data decompression circuit for a data stream having a repeated data word, the data stream including a series of data frames such that the repeated data word is removed from the series of data frames, each data frame including a header, comprising: a decompression engine configured to decompress each data frame into a corresponding decompressed data frame, the decompression engine being further configured to decode each header to identify whether word locations in the corresponding decompressed data frame should be filled with the repeated data word.
Another aspect of the invention relates to a method of decompressing a data frame into a decompressed data frame, comprising: decoding a header in the data frame to identify whether locations in the decompressed data frame correspond to a repeated data word and to identify whether locations in the decompressed data frame correspond to uncompressed data words.
Another aspect of the invention relates to a data decompression circuit, comprising: a data input circuit operable to receive a data stream that includes data frames having a header and body, the header indicating the location within the body of one or more data words absent from the body; and a data decompression engine coupled to the data input circuit, the data decompression engine operable to read the header of a data frame to determine the location within the body of an absent data word and to insert the absent word into its proper location.
Another aspect of the invention relates to a data frame for carrying compressed data, comprising: a body containing uncompressed data words; and
a header that indicates the location within the body of one or more data words absent from the body.
Use of the same reference symbols in different figures indicates similar or identical items.
During the configuration of a programmable logic device such as a field programmable gate array (FPGA), it is common that a substantial portion of the configuration data stream maps to logic resources that are not used. For example, an FPGA may contain thousands of logic blocks. However, when an FPGA is programmed to implement a user's desired logical function, only a subset of the available logic blocks is typically utilized. The configuration memory cells for these unused logical resources are programmed to a defined null state. For example, the configuration memory cells for an unused logic block may be programmed with all binary zeroes to signify that the logic block is to remain dormant during use of the corresponding programmable logic device. Alternatively, a null state may be indicated by having the configuration memory cells programmed with binary ones or some other binary combination. Thus, the configuration data stream for a programmable logic device will generally contain a substantial amount of repeated data words such as all binary zero (or all binary one) data words that correspond to the unused logical resources.
Other types of data streams will also include substantial amounts of repeated data words as well. The present invention exploits the redundant nature of such data streams by compressing the data streams into frames or groups of data such that the repeated data word is eliminated and thus absent from the groups. Each group of data includes a header that signifies the number and location of the absent data words corresponding to the original data stream prior to compression. After decompression, each group of data becomes a decompressed group of data comprising a number of data words. The number of data words corresponding to each decompressed frame/group of data may be varied according to individual design needs. Without loss of generality, the following discussion will assume that each decompressed frame corresponds to eight byte-long words. For such a frame size, eight bits may used in the header to signify whether a corresponding byte word in the decompressed frame is the repeated data word. A convenient correlation is that a first bit in the header corresponds to a first byte in the decompressed frame, a second bit in the header corresponds to a second byte in the decompressed frame, and so on. However, it will be appreciated that other mappings between bit positions in the header and corresponding words in the decompressed frame may be used.
An exemplary compression scheme may be described with respect to a programmable logic device. In a programmable logic device, the configuration data stream may contain many all-zero bytes corresponding to unused logical resources. A convenient compression scheme is to simply represent the all-zero bytes by a binary zero in the header. More generally, the data stream being compressed will contain some form of repeated data word. The header may represent this repeated data word by either a binary zero or a binary one. Without loss of generality, the following discussion will assume that this representation will be made using a binary zero.
An exemplary embodiment for a decompression circuit 100 for decompressing such a compressed data stream is illustrated in
The compressed data streams that may be decompressed by decompression circuit 100 may originate from a variety of sources. For example, if the device including decompression circuit 100 supports operation in a Joint Test Action Group (JTAG) mode, the corresponding JTAG interface may be used to couple a compressed serial bit stream to data process engine 110 as a hybrid JTAG input, TDI 140. Alternatively, data process engine 110 may receive a compressed serial bit stream as an input DIN 135. In a byte mode of operation, data process engine 110 may received a compressed byte stream as an input D[0:7] 155. To control whether data process engine 110 operates in a bit or byte mode (and also whether the bit mode is a hybrid JTAG bit mode), data process engine 110 may receive a control variable CFG[2:0] 160.
It will be appreciated that data process engine 110 is illustrated as receiving byte stream D[0:7} 155 for conceptual purposes only in that data process engine 110 need do nothing to assemble this incoming byte stream onto dbus_in 120. However, in bit mode, data process engine 110 will receive a bit at a time and will thus function to assemble received bits into bytes for dbus_in 120. Regardless of whether the operation occurs in a bit or byte mode, the various compressed data streams received as inputs TDI 140, DIN 135, and D[0:7] 155 will be synchronized according to corresponding clocks. For example, TDI 140 is synchronized according to a JTAG clock TCK 150. The remaining inputs are synchronized according a clock CCLK 145. In a bit mode of operation, clock CCLK 145 cycles at the bit rate whereas in a byte mode of operation, clock CCLK 145 cycles at the byte rate. Depending upon the mode of operation, data process engine 110 selects from these clocks to provide a mux_clock 125 to a decompression engine 170.
Operation of decompression circuit 100 may be better understood with reference to several example compressed data streams for dbus_in 120. In this exemplary embodiment, each data frame is of variable size that depends upon the number of uncompressed data bytes that follow the header. Each frame starts with a header followed by a body. The header indicates the number of uncompressed data bytes in the body. Because the header in this example will be assumed to have a length of one byte, the maximum number of uncompressed data bytes in each frame will be eight. However, it will be appreciated that different maximum numbers of data bytes will be supported by varying the header size. Decompression engine 170 includes a state machine or decoder that functions to decode the header and assemble a wide_data bus 175 accordingly. Thus, decompression engine 170 functions to receive a group of data, decode its header, and insert any required repeated data word into its proper location amongst the uncompressed words in the group of data to form a decompressed group of data corresponding to the received group of data. In the embodiment illustrated, a maximum number of eight uncompressed data bytes may follow each header for a given frame. Thus, the width of wide_data bus 175 is eight bytes. Consider the following example header b00110101 received on dbus_in 120 by decompression engine 170. Decompression engine 170 decodes this header and recognizes that four uncompressed data bytes will follow this header as determined by the binary ones in the header. These four uncompressed data bytes may be denoted as data words data through data. The position or order of these uncompressed data bytes in the decompressed group of data formed on wide_data bus 175 may be coded by their position in the header. Wide_data bus 175 may be considered as ordered from a least significant byte through a most significant byte. Given such an ordering and mapping it to corresponding bit positions in the header, decompression engine 170 may place data  as byte  on wide_data bus 175, data  will be byte , data  will be byte , and data  will be the byte .
The remaining bits in this example header are all zero. In this exemplary embodiment, zero bits correspond to the repeated data word in the compressed data stream, which is assumed to be an all zero byte. Thus, bytes , , , and  are all zero bytes. It will be appreciated, however, that the repeated data word need not equal an all zero word. The resulting decompression from the original frame on dbus_in 120 to the decompressed frame on wide_data bus 175 may be seen in the following representations:
If the original group of data (having a header and body) is:
Header Data Data Data Data b00110101 b10011110 b10010011 b1101110 b00100000
The resulting decompressed group of data is:
byte byte byte byte byte byte byte byte b00000000 b00000000 b10011110 b10010011 b00000000 b11101110 b00000000 b00100000
Given this form of decompression, it may be seen that if the header is all binary ones, the frame/group of data will comprise eight uncompressed data bytes. For example, suppose the group of data is:
header Data Data Data Data Data Data Data Data b11111111 b10011110 b10010011 b11101110 b00100000 b01111111 b10111111 b11011111 b11101111
Because there are no zero bits in this header, the resulting decompressed group of data is simply the eight bytes following the header:
On the other hand, if the header is all binary zeroes instead of binary ones as in the preceding example, the frame/group of data will simply comprise the header such that the group of data is represented as:
Given such a frame, the resulting decompressed data becomes:
Another exemplary frame may comprise just one uncompressed data byte that should map to the byte  position. Such a frame may be represented as:
header Data b00100000 b10011110
Given such a frame, the resulting decompressed data becomes:
When decompression engine 170 has decompressed a frame into the corresponding decompressed frame of data on wide_data bus 175, it may then signal that the data is ready for utilization using a signal such as data_rdy 180. In the embodiment illustrated, decompression circuit 100 is included in a programmable logic device such as an FPGA including a data shift register 185 that will receive the decompressed data from wide_data bus 175. When signaled by an assertion of data_rdy 180, data shift register 185 may receive the decompressed data on wide_data bus 175.
A more detailed block diagram of exemplary embodiments for data process engine 110 and decompression engine 170 is illustrated in
Decompression engine 170 may include a raw data process engine 220 that receives a header input 225 from header register 210. Raw data process engine 220 includes a state machine or processor configured to decode the received header as discussed previously. Because the header determines how many uncompressed data words will follow in the corresponding frame, the header also determines when the next header may be expected. A counter 230 may be triggered by header input 225 to set a count corresponding to the number of uncompressed data bytes that will follow the header. Raw data process engine 220 also decodes the header to determine the location of the uncompressed data bytes in the decompressed data frame being assembled on wide_data bus 175. To assemble the bytes onto wide_data bus 175, raw data process engine 220 may couple to byte registers 240, arranged from a least significant byte register 240 a to a most significant byte register 240 h. For example, if the header is b00010001, the binary ones in the header may be decoded to indicate that two uncompressed data bytes will follow in the body. The initial uncompressed data byte will be mapped to byte register 240 e whereas the subsequent uncompressed data byte will be mapped to byte register 240 a. Because all the remaining bits in this example header are zero, registers 240 b, 240 c, 240 d, 240 f, 240 g, and 240 h would be loaded with all zero byte words by raw data process engine 220. Registers 240 may all be clocked by mux_clock 125. To indicate which register should be enabled to respond to mux_clock 125 so as to register a byte provided by raw data process engine 220, raw data process engine 220 may provide a shift enable signal (not illustrated) to registers 240. Raw data process engine 220 may receive the uncompressed data bytes from either input bus D[0:7] 225 or shift register 205 depending upon whether a bit or byte mode of operation is enabled. Having loaded all registers 240, raw data process engine 220 may then activate data_ready signal 180 (
Data process engine 110 may include a multiplexer 250 to select for mux_clock 125. Multiplexer 250 receives clocks TCK 150 and CCLK 145 and selects the appropriate clock depending upon whether a hybrid JTAG mode of operation is enabled. The resulting mux_clock 125 may then be distributed to raw data process engine 220 and registers 240.
Advantageously, decompression engine 170 may process dbus_in 120 at a constant byte rate, regardless of the amount of compression involved. The timing of such a byte-by-byte constant processing may be determined by mux_clock 125. Should data process engine 110 be in bit mode, mux_clock 125 will cycle according to the bit rate. Thus, decompression engine 170 would simply respond to every eighth cycle of mux_clock 125 in the bit mode. In a byte mode of operation, decompression engine 170 may respond to every cycle of mux_clock 125 since it will cycle according to the byte rate in the byte mode of operation. Regardless of whether a byte or bit mode of operation is enabled, decompression engine 175 processes data on dbus_in 120 at a constant word rate, thereby eliminating the problems encountered in conventional decompression circuits regarding the different rates required for processing compressed and uncompressed data. It thus doesn't matter whether data on dbus_in 120 is compressed or uncompressed, compression engine 170 may process it a regular byte-by-byte basis. It will be appreciated that such processing may be applied to other word lengths besides bytes.
The regular processing of words by decompression engine 170 may be better understood with reference to the timing relationships illustrated in
It may be shown that the compression efficiency by the above-described compression scheme depends upon the frequency of the repeated data word that is being eliminated. Using the exemplary decompressed frame length of eight bytes, the overhead introduced by the header is 12.5%. Thus, so long as the repeated data word has a greater frequency than 12.5%, it would behoove a user to use the compression/decompression scheme described herein. In that regard, the embodiments of the present invention described herein are merely illustrative. For example, the repeated data word that is eliminated from each frame/group of data following a header need not be an all-zero data word. Moreover, the word size for the repeated data word and the uncompressed data words may be varied from the byte length described herein. Finally, the number of data words within each group of data need not be eight but instead may be varied as desired. Therefore, the above-described embodiments of the present invention are not limiting. It will thus be obvious to those skilled in the art that various changes and modifications may be made without departing from this invention in its broader aspects. Accordingly, the appended claims encompass all such changes and modifications as fall within the true spirit and scope of this invention.
|Patente citada||Fecha de presentación||Fecha de publicación||Solicitante||Título|
|US3717851||3 Mar 1971||20 Feb 1973||Ibm||Processing of compacted data|
|US4829475 *||1 Ago 1986||9 May 1989||Texas Instruments Incorporated||Method and apparatus for simultaneous address increment and memory write operations|
|US4899149||28 Feb 1986||6 Feb 1990||Gary Kahan||Method of and apparatus for decoding Huffman or variable-length coees|
|US5099481 *||28 Feb 1989||24 Mar 1992||Integrated Device Technology, Inc.||Registered RAM array with parallel and serial interface|
|US5173695||29 Jun 1990||22 Dic 1992||Bell Communications Research, Inc.||High-speed flexible variable-length-code decoder|
|US5467087 *||18 Dic 1992||14 Nov 1995||Apple Computer, Inc.||High speed lossless data compression system|
|US5563592 *||22 Nov 1993||8 Oct 1996||Altera Corporation||Programmable logic device having a compressed configuration file and associated decompression|
|US5745734||29 Sep 1995||28 Abr 1998||International Business Machines Corporation||Method and system for programming a gate array using a compressed configuration bit stream|
|US5754734||31 Oct 1995||19 May 1998||Motorola, Inc.||Method of transmitting voice coding information using cyclic redundancy check bits|
|US5768372||13 Mar 1996||16 Jun 1998||Altera Corporation||Method and apparatus for securing programming data of a programmable logic device|
|US5794033||24 Oct 1995||11 Ago 1998||International Business Machines Corporation||Method and system for in-site and on-line reprogramming of hardware logics with remote loading in a network device|
|US5805609 *||27 Jun 1997||8 Sep 1998||Samsung Electronics Co., Ltd.||Method and apparatus for testing a megacell in an ASIC using JTAG|
|US5872529||16 Jul 1997||16 Feb 1999||Altera Corporation||Dynamic datastream compression/decompression|
|US5953503||29 Oct 1997||14 Sep 1999||Digital Equipment Corporation||Compression protocol with multiple preset dictionaries|
|US6028445||30 Dic 1997||22 Feb 2000||Xilinx, Inc.||Decoder structure and method for FPGA configuration|
|US6038185||12 May 1998||14 Mar 2000||Atmel Corporation||Method and apparatus for a serial access memory|
|US6044025||4 Feb 1999||28 Mar 2000||Xilinx, Inc.||PROM with built-in JTAG capability for configuring FPGAs|
|US6127953 *||23 Mar 1998||3 Oct 2000||Siemens Information And Communication Networks, Inc.||Apparatus and method for compressing data containing repetitive patterns|
|US6191710 *||30 Dic 1998||20 Feb 2001||Intel Corp.||Data compression and decompression method and system for data compression and decompression|
|US6327634||25 Ago 1998||4 Dic 2001||Xilinx, Inc.||System and method for compressing and decompressing configuration data for an FPGA|
|US6483342||25 May 2001||19 Nov 2002||Lattice Semiconductor Corporation||Multi-master multi-slave system bus in a field programmable gate array (FPGA)|
|US6493862||25 Jul 2000||10 Dic 2002||Xilinx Inc.||Method for compressing an FPGA bitsream|
|US6507214||18 Jul 2001||14 Ene 2003||Cypress Semiconductor Corporation||Digital configurable macro architecture|
|US6507943||26 Sep 2000||14 Ene 2003||Xilinx, Inc.||Method of compressing a bitstream of an FPGA|
|US6525678||5 Oct 2001||25 Feb 2003||Altera Corporation||Configuring a programmable logic device|
|US6563437||2 Oct 2000||13 May 2003||Cypress Semiconductor Corporation||Method and apparatus for using programmable logic device (PLD) logic for decompression of configuration data|
|US6564285||14 Jun 2000||13 May 2003||Intel Corporation||Synchronous interface for a nonvolatile memory|
|US6744388 *||19 Jun 2002||1 Jun 2004||Xilinx, Inc.||Hardware-friendly general purpose data compression/decompression algorithm|
|US6772230||25 May 2001||3 Ago 2004||Lattice Semiconductor Corp.||Field programmable gate array (FPGA) bit stream format|
|US6785165||4 Dic 2002||31 Ago 2004||Renesas Technology Corporation||Semiconductor device|
|US6885227||29 Jul 2003||26 Abr 2005||Lattice Semiconductor Corporation||Clock generator with skew control|
|US6903574||29 Jul 2003||7 Jun 2005||Lattice Semiconductor Corporation||Memory access via serial memory interface|
|US7019577||29 Jul 2003||28 Mar 2006||Lattice Semiconductor Corporation||Clock generator|
|US7034599||27 Ene 2005||25 Abr 2006||Lattice Semiconductor Corporation||Clock generator with skew control|
|US20020091905 *||11 Ene 2002||11 Jul 2002||Interactive Silicon, Incorporated,||Parallel compression and decompression system and method having multiple parallel compression and decompression engines|
|US20040061147||19 Ene 2001||1 Abr 2004||Ryo Fujita||Electronic circuit device|
|US20040064622||30 Sep 2002||1 Abr 2004||Smith Winthrop W.||Signal processing resource with sample-by-sample selective characteristics|
|US20050229061 *||1 Abr 2004||13 Oct 2005||Sung-Ho Kang||Method of efficiently compressing and decompressing test data using input reduction|
|1||Altera Corporation, Configuring Cyclone II Devices, Cyclone Device Handbook, Jul. 2005, vol. 1, Chapters 13-14, pp. 13-1 to 13-4.|
|2||Atmel Corporation, AT17LV(A) Series FPGA Configuration Memory, Ref. 0437K-CNFG-May 2003, pp. 1-18.|
|3||Atmel Corporation, FPGA Configuration EEPROM Memory, 2322F-CNFG, Jun. 2005, pp. 1-18.|
|4||Booth, U.S. Appl. No. 11/494,862, filed Jul. 28, 2006.|
|5||Callahan, U.S. Appl. No. 11/530,620, filed Sep. 11, 2006.|
|6||Chen, U.S. Appl. No. 11/346,817, filed Feb. 3, 2006.|
|7||Tang, U.S. Appl. No. 10/809,658, filed Mar. 25, 2004.|
|8||Tang, U.S. Appl. No. 11/243,255, filed Oct. 4, 2005.|
|Patente citante||Fecha de presentación||Fecha de publicación||Solicitante||Título|
|US7808405||30 Mar 2009||5 Oct 2010||Lattice Semiconductor Corporation||Efficient bitstream compression|
|US7902865||15 Nov 2007||8 Mar 2011||Lattice Semiconductor Corporation||Compression and decompression of configuration data using repeated data frames|
|US8780940 *||6 Ago 2009||15 Jul 2014||Electronics And Telecommunications Research Institute||Method and apparatus for compressing frame|
|US20120002683 *||6 Ago 2009||5 Ene 2012||Sung-Jin You||Method and apparatus for compressing frame|
|US20120173846 *||14 Dic 2011||5 Jul 2012||Stmicroelectronics (Beijing) R&D Co., Ltd.||Method to reduce the energy cost of network-on-chip systems|
|US20130125186 *||19 Jul 2011||16 May 2013||Lg Electronics Inc||Method for transmitting/receiving media and device for transmitting/receiving using same|
|CN102567277A *||30 Dic 2010||11 Jul 2012||世意法（北京）半导体研发有限责任公司||Method for reducing power consumption through network-on-chip system|
|WO2012174807A1 *||30 Sep 2011||27 Dic 2012||Shenzhen State Microelectronics Co., Ltd.||Method and system for compressing and decompressing configuration data stream of programmable logic device|
|Clasificación de EE.UU.||341/87, 326/37, 341/67, 326/38, 341/65, 341/50, 707/999.101|
|Clasificación cooperativa||Y10S707/99942, H03M7/3086|
|28 Feb 2005||AS||Assignment|
Owner name: LATTICE SEMICONDUCTOR CORPORATION, OREGON
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:MA, BENNY;KOW, SAN-TA;WU, ANN;AND OTHERS;REEL/FRAME:015801/0928
Effective date: 20050207
|21 Sep 2010||CC||Certificate of correction|
|13 Feb 2013||FPAY||Fee payment|
Year of fee payment: 4
|18 Mar 2015||AS||Assignment|
Owner name: JEFFERIES FINANCE LLC, NEW YORK
Free format text: SECURITY INTEREST;ASSIGNORS:LATTICE SEMICONDUCTOR CORPORATION;SIBEAM, INC.;SILICON IMAGE, INC.;AND OTHERS;REEL/FRAME:035223/0001
Effective date: 20150310